二进神经网络的最稳健设计

本文基于对ｎ维超立方体的分割超平面应处在任两个不同类相邻顶点连线的中点、且与连 的思想，设计了能够实现逻辑功能的具有次强和最强稳健能力的我层前向神经网络、给出了稳健神经元之间的连接权和偏置值的设计规则，并服运用这一规则能够设计出既能实现所希望逻辑功能，又具有对输入数据的次强和最强稳健能力的前向网络，相应的网络规模分别为ｎ－ｋ－１和ｎ－ｎ－ｋ－１由于所构造的网络其连接权值只可能是＋１、－１、０，使得     

二进联想Hopfield神经网络的稳定点研究

稳定点是决定人工神经网络联想记忆能力的最重要因素。本文在二进联想Hopfield网络的稳定点统计分布、计数和快速判别等方面给出了若干新结果。     

采用BP神经网络的智能抗干扰决策引擎研究

在认知抗干扰通信系统中,智能决策是其核心,根据干扰环境,对系统的干扰抑制方式、频谱资源分配、调制编码方式和功率调整信息进行最优决策。现有的抗干扰通信系统的智能决策多采用遗传算法、人工蜂群算法等,面对日益复杂的电磁环境,通常这些算法不具有对新干扰的泛化能力。BP神经网络算法简单、具有一定的容错能力和泛化能力,本文设计并分析了一种基于BP神经网络的抗干扰实时决策引擎模型,根据系统性能设计了输入输出数据的预处理方式和判别标准,阐述了决策实现步骤,分析了算法参数;通过系统性能仿真,验证了文中提出的实时决策引擎的强抗干扰性能。与采用遗传算法和人工蜂群算法的决策引擎相比,本文提出的决策引擎决策速度更快且具有泛化能力和容错能力。      

Distributed Restoration in Optical Networks using Feed-forward Neural Networks

A new method is proposed for determining protection paths in an optical network where users have different characteristics in terms of reliability needs and security restrictions. Survivability is achieved by distributed mesh protection. Over the preplanned primary and backup capacity, optimal routing and wavelength assignment is carried out. In case of a network failure, protection routes and optimum flow values on these protection routes are extracted from a previously trained feed-forward neural network which is distributed over the optical data communications network.     

NN-EdgeBuilder：面向边缘端设备的高性能神经网络推理框架

飞速发展的神经网络已经在目标检测等领域取得了巨大的成功,通过神经网络推理框架将网络模型高效地自动部署在各类边缘端设备上是目前重要的研究方向。针对以上问题,该文设计一个针对边缘端FPGA的神经网络推理框架NN-EdgeBuilder,能够利用基于多目标贝叶斯优化的设计空间探索算法充分探索网络每层的并行度因子和量化位宽,接着调用高性能且通用的硬件加速算子来生成低延迟、低功耗的神经网络加速器。该文使用NN-EdgeBuilder在Ultra96-V2 FPGA上部署了UltraNet和VGG网络,生成的UltraNet-P1加速器与最先进的UltraNet定制加速器相比,功耗和能效比表现分别提升了17.71%和21.54%。与主流的推理框架相比,NN-EdgeBuilder生成的VGG加速器能效比提升了4.40倍,数字信号处理器(DSP)的计算效率提升了50.65%。     

Call Admission Control Decision Maker Based on Optimized Fuzzy Inference System for 5G Cloud Radio Access Networks

Wireless Personal Communications - In C-RAN research, different functional splits are considered for 5G mobile crosshaul networks. The typical fronthaul network based on optical fibers will be...     

Rapid and Precise GLONASS GDOP Approximation using Neural Networks

Russian global navigation satellite system (GLONASS) provides civilian and military users three-dimensional position determination and navigation services as same as US global positioning system. Geometric dilution of precision (GDOP) provides a simple interpretation of positioning precision. Usual method for GLONASS GDOP calculation is matrix inversion. However this process imposes a huge calculation load on receiver, especially when large number of visible satellites exists. To overcome this problem, artificial neural network is used. Different configurations and training methods are simulated on a data base obtained by a GLONASS receiver. Then navigation precision and execution times are explored and compared. Results show that recurrent neural network has 0.00024 RMS error, which is the best against other focused tools including feed forward back propagation and radial basis function neural network with usual training and with genetic algorithm adopted weights and biases.     

An efficient memristor crossbar architecture for mapping Boolean functions using Binary Decision Diagrams (BDD)

The memristor is considered as the fourth fundamental circuit element along with resistor, capacitor and inductor. It is a two-terminal passive circuit element whose resistance value changes based on the amount of charge flowing through it. Another property of the memristor is that its resistance change is non-volatile in nature, and hence can be used for non-volatile memory applications. Researchers have been exploring memristors from various perspectives such as logic design and storage applications. In this paper, a slicing crossbar architecture for the efficient mapping of Boolean functions is proposed which exploits gate level parallelism using the memristor aided logic (MAGIC) design style. A Boolean function is first represented as a Binary Decision Diagram (BDD). The BDD nodes are expressed as netlists of NOR and NOT gates, and are mapped to the proposed slicing crossbar architecture with parallel node evaluation where possible. This is the first approach that combines BDD-based synthesis with MAGIC gate evaluation on memristor crossbar, while at the same time avoiding crossbar-related problems using a slicing architecture. Experimental evaluations on standard benchmark functions show considerable improvement in the solutions.     

Protocol for Dynamic Ad-Hoc Networks Using Distributed Spanning Trees

A dynamic ad-hoc network consists of a collection of mobile hosts with frequently changing network topology. We propose a distributed algorithm that adapts to the topology by utilizing spanning trees in the regions where the topology is stable, and resorting to an intelligent flooding-like approach in highly dynamic regions of the network. Routing is performed using the spanning trees based a hold-and-forward or shuttling mechanisms. We introduce the notion of connectivity-through-time and the parameter holding-time as new fundamental concepts that can be used by ad-hoc routing algorithms. For various network connectivity scenarios we evaluate the impact of these concepts on the performance of ad-hoc routing algorithms. Using simulation, we study the throughput, reachability and message–reachability ratio of the proposed schemes under various connection/disconnection rates and holding times.     

E-BEB: Enhanced Binary Exponential Backoff Algorithm for Multi-hop Wireless Ad-hoc Networks

Binary exponential backoff algorithm is the de-facto medium access control protocol for wireless local area networks, and it has been employed as the standard contention resolution algorithm in multi-hop wireless ad-hoc networks. However, this algorithm does not function well in multi-hop wireless environments due to its several performance issues and technical limitations. In this paper, we propose a simple, efficient, priority provision, and well performed contention resolution algorithm called enhanced binary exponential backoff (E-BEB) algorithm for impartial channel access in multi-hop wireless ad-hoc networks. We also provide a simple and accurate analytical model to study the system saturation throughput of the proposed scheme. Simulations are conducted to evaluate the performance of E-BEB algorithm. The results show that the E-BEB algorithm can alleviate the fairness problem and support multimedia transmission in multi-hop wireless environments.     

前向神经网络动态学习

在网络中同一隐层的所有神经元对不同样本的输出所构成的向量相应线性无关、本文利用这一基本事实，对每一隐层引入了一相关向量及相应的无关度，根据无关度对该隐层神经元数目进行删除或增加，同时适当调整相应的网络权值，这样做既可以避免对隐层神经元的预先确定，同时还可以在学习过程中逃离局部极小。根据删除神经元对网络所带来的误差的详细分析，给出了确定删除隐层神经元的最估规则，数据实验表明了该方法的有效性。     

Weighted Learning for Feedforward Neural Networks

In this paper, we propose two weighted learning methods for the construction of single hidden layer feedforward neural networks. Both methods incorporate weighted least squares. Our idea is to allow the training instances nearer to the query to offer bigger contributions to the estimated output. By minimizing the weighted mean square error function, optimal networks can be obtained. The results of a number of experiments demonstrate the effectiveness of our proposed methods.     

Energy Scaling Laws for Distributed Inference in Random Fusion Networks

The energy scaling laws of multihop data fusion networks for distributed inference are considered. The fusion network consists of randomly located sensors distributed i.i.d. according to a general spatial distribution in an expanding region. Under Markov random field (MRF) hypotheses, among the class of data-fusion policies which enable optimal statistical inference at the fusion center using all the sensor measurements, the policy with the minimum average energy consumption is bounded below by the average energy of fusion along the minimum spanning tree, and above by a suboptimal policy, referred to as Data Fusion for Markov Random Fields (DFMRF). Scaling laws are derived for the energy consumption of the optimal and suboptimal fusion policies. It is shown that the average asymptotic energy of the DFMRF scheme is strictly finite for a class of MRF models with Euclidean stabilizing dependency graphs.     

Multicast Routing in Wireless Mesh Networks: Minimum Cost Trees or Shortest Path Trees?

There exist two fundamental approaches to multicast routing: shortest path trees (SPTs) and minimum cost trees (MCTs). The SPT algorithms minimize the distance (or cost) from the sender to each receiver, whereas the MCT algorithms minimize the overall cost of the multicast tree. Due to the very large scale and unknown topology of the Internet, computing MCTs for multicast routing in the Internet is a very complex problem. As a result, the SPT approach is the more commonly used method for multicast routing in the Internet, because it is easy to implement and gives minimum delay from the sender to each receiver, a property favored by many real-life applications. Unlike the Internet, a wireless mesh network (WMN) has a much smaller size, and its topology can be made known to all nodes in the network. This makes the MCT approach an equally viable candidate for multicast routing in WMNs. However, it is not clear how the two types of trees compare when used in WMNs. In this article we present a simulation-based performance comparison of SPTs and MCTs in WMNs, using performance metrics, such as packet delivery ratio, end-to-end delay, and traffic impacts on unicast flows in the same network.     

A Logarithmic Binary Decision Diagram-Based Method for Multistate System Analysis

Multi-state systems (MSS) are systems in which both the systems, and/or their components may exhibit multiple performance levels or states. MSS can model complex behaviors such as shared loads, performance degradation, imperfect fault coverage, standby redundancy, and limited repair resources. The non-binary state property of MSS, and their components makes the analysis of MSS challenging. In this paper, we propose efficient logarithmically-encoded binary decision diagram (LBDD)-based methods for analysing MSS. The application and advantages of the proposed LBDD-based approaches, as compared to the existing binary decision diagram-based approaches, are demonstrated through the analyses of practical MSS examples, and a set of benchmark examples.      

结合二叉判决图和布尔可满足性的等价性验证算法

本文提出了一种结合二叉判决图BDD和布尔可满足性SAT的新颖组合电路等价性验证技术.算法是在与/非图AIG中进行推理,并交替使用BDD扩展和基于电路SAT解算器简化电路.如尚未解决,将用基于合取范式SAT解算器进行推理.与已有算法相比主要有如下改进:在AIG中结合多种引擎进行简化,不存在误判可能;充分利用了基于电路解算器和基于合取范式解算器各自优点,减小了SAT推理的搜索空间.实验结果表明了本算法的有效性.     

Group Testing for Binary Markov Sources: Data-Driven Group Queries for Cooperative Sensor Networks

Group testing has been used in many applications to efficiently identify rare events in a large population. In this paper, the concept of group testing is generalized to applications with correlated source models to derive scheduling policies for sensors' adopting cooperative transmissions. The tenet of our work is that in a wireless sensor network it is advantageous to allocate the same channel dimensions to all sensor sources that have the same response to a sequence of queries or tests. That is, nodes that have the same data attributes should transmit as a cooperative super-source. Specifically, we consider the case where sensors' data are modeled spatially as a one-dimensional Markov chain. Two strategies are considered: the recursive algorithm and the tree-based algorithm. The recursive scheme allows us to illustrate the performance of group testing for finite populations while the tree-based algorithm is used to derive the achievable scaling performances of the class of group testing strategies as the number of sensors increases. We show that the total number of queries required to gather all sensors' data scales in the order of the joint entropy. A further generalization of this concept provides the basis of deriving efficient data-gathering algorithms for correlated sources.